Digital fix for wow and flutter

There was a feature on National Public Radio's All Things Considered program yesterday that demonstrated some before/after examples of how old movie soundtracks are being restored to correct wow and flutter.  A bit short on technical details, but interesting anyway.  Here's the link: http://www.npr.org/templates/story/story.php?storyId=7489316

Mod note:  Link changed so that it works

Read all about it here - http://www.plangentprocesses.com - or, at least as much as anyone is going to tell you about it .

.
Well, they are going to have to oversample pretty substantially if they expect to "sync" on the
record bias tone, typically 60-120Khz on better recorders.  This implies a sample rate of 240Khz
to digitize the higher frequency.  Assuming, of course that any tones of this frequency actually
make if out of the playback electronics!! 


Hmmmm....
.

They could play it back at half or quarter speed perhaps.

They could play it back at half or quarter speed perhaps.

That works fine, and certainly has to be the way that this is to be done, if you are doing it for real. The bias frequency is way above the extinction frequency of the replay head otherwise. In the AES paper it goes on about 'special high frequency heads', and I suppose that if you fitted a video replay head to an ATR you might be able to reproduce the bias and heterodyne it down to a useable frequency, but this seems to be altogether more hard work than digitising it at a much slower rate. It doesn't matter about the added wow and flutter you might introduce - because you're going to remove it anyway!

.

They could play it back at half or quarter speed perhaps.

Ah yes, that would help considerably, but now the problem becomes the LF as the Low Frequency limits of the head are also pretty
steep and bounded by the down track pole length as much the gap length.  P.S: the HF rolloff is hypercylic so it IS possible to get
output above the first null frequency determined by the gap length if your are lucky and the actual bias frequency falls near a
hypercyclic peak.  However the bias suppression filter and general rolloff of the playback electronics would likely prevent this HF from
passing thru anyway.

Still slowing down is a much better idea.  I will try this with some of my old tapes and see if I can pickup some of the bias tone.
Don't think I will try the W&F removal, I'm just curious about the tone recovery. It would be easy to design a PLL that would track
this and yield a correction signal.
.

Is there any way this could work on cassette?  Interesting that all but one of the restorations came from reels.  The exception was a shrunken acetate film -- which would still work if it's 35mm mag.  All of these formats have a relatively fast speed, which presumably does a better job of preserving bias remnants than the micro-tiny wave patterns placed on a cassette. 

By the way, I can't get the Plangent website to open.

Ah yes, that would help considerably, but now the problem becomes the LF as the Low Frequency limits of the head are also pretty steep and bounded by the down track pole length as much the gap length.  P.S: the HF rolloff is hypercylic so it IS possible to get output above the first null frequency determined by the gap length if your are lucky and the actual bias frequency falls near a hypercyclic peak.  However the bias suppression filter and general rolloff of the playback electronics would likely prevent this HF from passing thru anyway.

Apart from all of that, it gets complicated because of how the bias frequency is actually imprinted onto the tape. Being at a relatively high frequency, it only gets recorded onto the surface layer of oxide - it really doesn't penetrate the layer much at all. Consequently it's quite vulnerable. AS far as LF response is concerned, I think that you stand a much better chance of reproducing at least some of this than you do of relying on the bias falling into one of the extended response peaks in a replay head. The response in those peaks reduces at quite a rate, one way or another, simply because of the iron losses in the average replay head - that's why they have to use ferrites in video heads.

But most tape machines can (with some care) reproduce 12.5Hz or thereabouts - so even if you ran the tape at quarter-speed, you'd still have a pretty good response down to 50Hz when you speeded up the results. I must admit that I haven't tried this as an experiment, but I do know that if you slow 15ips recorded tape down to 1 7/8, you can hear most bias signals quite clearly, so reproducing them at 1/4 speed should be quite feasible, I think - even if you have to disconnect the bias filter!

By the way, I can't get the Plangent website to open.

Neither can I - now.  The link is correct, so I assume their server is down.

Too many of us showing an interest?

Too many of us showing an interest?

Possibly - I just checked and it's up again.

As far as I can make out, the main disadvantage for those of us who might be interested is that the technology is not be sold.  You have to send them your tracks (and pay the fee to have the work done, of course).

[restoration by deconvolution]

It was an interesting piece on NPR.
I caught it too, but wondered why they devoted so much time to it (it was a long piece, compared to most).

In regard to the technique itself though, I wondered to myself at the time if it didn't roughly correspond to the technique of restoration by deconvolution applied to images (I'm asking), perhaps with some novel twists? Wow and flutter wouldn't correspond exactly to gaussian or circular blurring per se, but it would to motion blurring, which prompted me to wonder. I suppose a key difference is that no component of the signal of interest is sampled, but rather the bias, which would seemingly have no correlary counterpart in deconvolution as applied with images.

On second thought, maybe it corresponds more closely to "unshake" algos applied to videos...

On second thought, maybe it corresponds more closely to "unshake" algos applied to videos...

Not so sure about that... Part of the problem with audio is that if you use the material itself, and differential changes in it, to establish a degree of correction you will never know quite where you are, and whether what you thought was a fixed entity actually is or not. It's different from motion blur or camera shake, because in this you have a fixed entity that has shifted in position, either vertically or horizontally - but still it has remained essentially the object it was - just shifted. The real problem with wow and flutter is that on a video signal, you couldn't detect it at all - because it would represent minute changes in the time frame, and that's the one thing you can't see in relation to individual frames, because persistence of vision won't allow us to detect this at all above about 12 fps.

On a lot of older equipment like 16mm mag/opt projectors, the speed of the picture presentation could alter quite a bit before anybody noticed, as long as these alterations were smoothed out before the film hit the sound sensor. That's the purpose of the 'loop' that causes so much hassle - it lets a flywheel smooth out jerkiness in the film motion before it hits the sound head. The only way you could tell that it was running at slightly the wrong speed is if everything sounds wrong (there's a distinct limit to how much smoothness you can restore before the loop tightens)

But the upshot of all this is that you can't really rely on the content at all for speed determination - it has to be something independent of this, which is where the bias comes in.

...which is to say, this probably can't begin to work on cassettes that lack either a bias-frequency recording or sprocket holes (16/35mm mag).  I take it to mean that without an externally applicable time reference, the Plangent approach is not possible.